ALPBench: A Benchmark for Attribution-level Long-term Personal Behavior Understanding

Lu Ren 1,* , Junda She 1,* , Xinchen Luo 1,* , Tao Wang 1,* , Xin Ye 1 , Xu Zhang 1 , Muxuan Wang 1 , Xiao Yang 1 , Chenguang Wang 1 , Fei Xie 1 , Yiwei Zhou 1 , Danjun Wu 1 , Guodong Zhang 1 , Yifei Hu 1 , Guoying Zheng 1 , Shujie Yang 1 , Xingmei Wang 1 , Shiyao Wang 1 , Yukun Zhou 1 , Fan Yang 1 · 2026

Read ALPBench: A Benchmark for Attribution-level Long-term Personal Behavior Understanding by Lu Ren 1,* , Junda She 1,* , Xinchen Luo 1,* , Tao Wang 1,* , Xin Ye 1 , Xu Zhang 1 , Muxuan Wang 1 , Xiao Yang 1 , Chenguang Wang 1 , Fei Xie 1 , Yiwei Zhou 1 , Danjun Wu 1 , Guodong Zhang 1 , Yifei Hu 1 , Guoying Zheng 1 , Shujie Yang 1 , Xingmei Wang 1 , Shiyao Wang 1 , Yukun Zhou 1 , Fan Yang 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03056.txt.

Description

Predicting what a person will like from their long history is benchmarked by Lu Ren and 25 co-authors in a 2026 paper. Language models could personalize recommendations by thinking over past actions. Item-focused benchmarks, however, cannot judge preferences for items that are entirely new, and short-term requests do not reflect enduring interests that persist across years of activity.

Their benchmark, ALPBench, predicts combinations of attributes a user will find interesting instead of given items. Ground truth then exists even for newly introduced items. Histories are written as natural-language sequences, supporting interpretable thinking. The task remains highly challenging for current models. It requires capturing complex interactions among many attributes and thinking over long actions sequences at once.

Catalog record

CREATORLu Ren 1,* , Junda She 1,* , Xinchen Luo 1,* , Tao Wang 1,* , Xin Ye 1 , Xu Zhang 1 , Muxuan Wang 1 , Xiao Yang 1 , Chenguang Wang 1 , Fei Xie 1 , Yiwei Zhou 1 , Danjun Wu 1 , Guodong Zhang 1 , Yifei Hu 1 , Guoying Zheng 1 , Shujie Yang 1 , Xingmei Wang 1 , Shiyao Wang 1 , Yukun Zhou 1 , Fan Yang 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03056

Browse the full library · About the Tower of Babel